The Full Transformer — Encoder + Decoder

Attention is the star. Everything else — residuals, normalization, feed-forward, cross-attention — is the scaffolding that lets you stack it deep.